Proposes SPOKE, a policy optimization method for poker agents
What happened
Researcher Phurinat Polasa proposes SPOKE (Strategic Preference Optimization with KL-anchored Equilibrium), a method for training agents to make strategic decisions in poker. It combines a teacher-relative value penalty, policy anchoring, risk constraints, and structured rationale supervision, and is compared against SFT, DPO, GRPO, and GSPO. The study reports that across three runs using SmallThinker-4B-A0.6B as the policy, SPOKE achieved the lowest average exploitability in both poker variants, Kuhn (0.213) and Leduc (1.081), and the highest top-1 agreement rate on the six-player table (0.775). These are the experimental results self-reported by the study.
AI-written from the coverage · updated 2026-10-21
Coverage timeline
Follow the coverage to see the story from different sides.
- arXiv Board & Classic Game AITeaching AI Agent to Play Poker Strategically Through Preference Optimization
Study proposes SPOKE, a strategic preference optimization method for training poker agents' strategic decisions, and compares it with SFT, DPO, GRPO, and GSPO. Across three runs with the SmallThinker-4B-A0.6B policy, SPOKE had the lowest average exploitability in Kuhn (0.213) and Leduc (1.081), and the highest top-1 agreement rate on the six-player table (0.775).